Cortical surface segmentation and mapping

Duygu Tosun, Maryam E. Rettmann, Xiao Han, Xiaodong Tao, Chenyang Xu, Susan M. Resnick, Dzung L. Pham, Jerry L. PrinceNeuroImage 2004 · DOI

TL;DR

This paper presents CRUISE, a comprehensive, near-automatic approach for reconstructing, flattening, and segmenting the human cerebral cortex from MRI data. This framework enables detailed analysis of cortical thickness and sulcal patterns, crucial for understanding brain function, development, and disease.

What Problem Does This Paper Address?

Accurately mapping the human cerebral cortex is fundamental for neuroscience and medicine. This involves reconstructing its complex 3D surface, which is essential for various analyses like measuring cortical thickness, identifying functional areas, and planning neurosurgery. Traditional methods often struggle with the intricate folds (gyri and sulci) of the cortex, imaging artifacts (noise, intensity inhomogeneities), and the need for computationally efficient and robust pipelines. Specifically, precisely delineating the boundaries of the gray matter (GM) and its relationship with white matter (WM) and cerebrospinal fluid (CSF) is challenging due to partial volume effects. Furthermore, comparing cortices across individuals requires standardized representations, such as flattened or spherical maps, and reliable methods for segmenting key anatomical features like sulci are needed for registration and analysis.

Methods

The core of the approach is the CRUISE (Cortical Reconstruction Using Implicit Surface Evolution) method. It aims to find three nested surfaces: the inner (GM/WM), pial (GM/CSF), and central (geometric center) surfaces.

Key steps and novel contributions include:

  • Fuzzy Tissue Classification (FANTASM): Instead of hard segmentation, FANTASM uses a "soft" classification to retain more information and robustly handle noise and intensity inhomogeneities.
  • Automatic Editing (AutoFill): This algorithm automatically fills ventricles and subcortical GM structures within the WM membership function, refining the inner cortical surface estimate.
  • Topology Correction (GTCA): A graph-based algorithm ensures the extracted WM isosurface has a topology consistent with a sphere, removing artifacts like "handles."
  • Anatomically Consistent Enhancement (ACE): This procedure modifies the GM segmentation to better delineate sulci, especially in cases of "back-to-back" sulcal banks, by using a weighted distance skeleton that favors CSF regions.
  • Nested Surface Reconstruction: A topology-preserving geometric deformable model (TGDM) is used iteratively to reconstruct the inner, central, and pial surfaces, ensuring proper nesting.
  • Surface Inflation and Spherical Mapping: A standardized partially flattened cortex (PFC) is created through parametric relaxation and iterated closest point registration. This PFC is then mapped to a sphere, providing a standardized coordinate system for visualization and registration.
  • Sulcal Segmentation: A multi-step process involving sulcal/gyral classification, geodesic depth calculation (using a fast marching method), watershed algorithm, and catchment basin merging is used to accurately segment sulci.

Datasets

The paper utilizes magnetic resonance (MR) images from human cerebral cortices. For the study on cortical thickness changes, they analyzed 105 data sets from 35 individuals obtained from the Baltimore Longitudinal Study of Aging (BLSA) database, with scans taken at three different time points over a 4-year interval.

Key Insights and Ideas

  • Fuzzy classification is crucial: Moving beyond hard segmentation with FANTASM significantly improves robustness to noise and partial volume effects, which are pervasive in MRI.
  • Topology matters: Ensuring a topologically correct surface (like a sphere) is essential for reliable cortical reconstruction and subsequent analyses. GTCA addresses this critical issue.
  • Sulci are challenging but important: The ACE method highlights the difficulty in accurately segmenting sulci due to their complex geometry and partial volume effects, and the proposed solution is key for detailed cortical analysis.
  • Standardization is key for comparison: Creating standardized representations like the Partially Flattened Cortex (PFC) and spherical maps is vital for comparing brain structures and functions across individuals and for building atlases.
  • Cortical thickness changes with age: The study provides evidence that cortical thickness decreases with age, with sulcal regions showing a slightly more pronounced thinning.

Key Results

  • The CRUISE framework provides a near-fully automatic and computationally fast approach for cortical reconstruction, flattening, and sulcal segmentation.
  • The nested surface reconstruction successfully generates anatomically feasible inner, central, and pial cortical surfaces.
  • The sulcal segmentation method effectively identifies distinct regions corresponding to cortical folds.
  • The study on the BLSA dataset found a significant negative correlation between age and mean cortical thickness across global, sulcal, and gyral regions, indicating cortical thinning with age.
  • Mean sulcal thickness was consistently found to be less than mean gyral thickness.

Limitations and Future Work

  • Longitudinal analysis limitations: The paper notes that the longitudinal analysis of cortical thickness changes over the 4-year interval did not yield significant results, possibly due to the small sample size and short time span. They plan to investigate this further with a larger dataset and longer follow-up periods.
  • Computational complexity: While described as computationally fast, the multi-step nature of the pipeline suggests that significant computational resources are still required.
  • Generalizability: While the approach is presented as comprehensive, the paper doesn't explicitly detail its performance on diverse populations or across different MRI scanner types, which could be a future area of investigation.
  • Future directions: The authors suggest that their approach can be used for answering questions about cortical thickness changes and other morphometric changes associated with aging. The developed methods are also valuable for creating atlases and for studying brain development and disease.

Summary updated

BibTeX
@inproceedings{tosun2004cortical,
  author = {Duygu Tosun and Maryam E. Rettmann and Xiao Han and Xiaodong Tao and Chenyang Xu and Susan M. Resnick and Dzung L. Pham and Jerry L. Prince},
  title = {{Cortical surface segmentation and mapping}},
  booktitle = {NeuroImage},
  year = {2004},
  doi = {10.1016/j.neuroimage.2004.07.042},
}
Figure